Bayesian regression analysis of data with random effects covariates from nonlinear longitudinal measurements

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Bayesian Inference

Background:

• Joint models link longitudinal measurements with primary response variables.
• Existing methods using nonlinear mixed-effects models face computational challenges.
• Understanding disease progression and treatment efficacy often requires analyzing joint data.

Purpose of the Study:

• To develop a Bayesian estimation procedure for joint models with nonlinear mixed-effects longitudinal data.
• To address computational efficiency issues in existing likelihood approximation strategies.
• To analyze the association between longitudinal hormone levels and pregnancy outcomes.

Main Methods:

• Implemented a Bayesian estimation procedure for joint models.
• Utilized a nonlinear mixed-effects model for longitudinal trajectories.
• Incorporated a generalized linear model for the primary response.
• Developed a prior structure enabling Markov Chain Monte Carlo (MCMC) sampling.
• Accounted for correlated errors in the longitudinal model.

Main Results:

• The Bayesian approach offers computational efficiency for joint models.
• The method successfully predicted normal versus abnormal pregnancy outcomes using early hormone levels.
• Modeling correlated errors is crucial for accurate analysis and avoiding misspecification.

Conclusions:

• The proposed Bayesian joint model provides an efficient and flexible framework for analyzing complex correlated data.
• Accurate modeling of longitudinal data, including correlated errors, is essential for reliable prediction of clinical outcomes.
• This approach has significant implications for understanding disease mechanisms and evaluating therapeutic interventions.